夜晚视频目标检测中的车辆灯光干扰消除方法

针对夜晚视频目标检测中的车辆灯光干扰问题，提出一种基于支持向量机(SVM)的灯光干扰消除方法。用背景差方法对输入视频进行初始分割得到前景区域，把前景区域划分为子块，提取子块的灰度分布特征和纹理特征；选择一个最佳特征子集作为输入向量，对SVM分类器进行训练识别；将识别为灯光的子块去除。实验结果表明，与其他分类器相比，基于SVM的方法在准确率和实时性方面性能较优。

Elimination Method of Vehicle Light Interference in Nighttime Video Object Detection

To solve the problem of vehicle light interference on nighttime object detection, a method of eliminating the light interference based on Support Vector Machine（SVM） is proposed in this paper. Foreground region is extracted from the input video by means of background subtraction. Features of gray distribution and texture are extracted from the blocks, which foreground region is divided into. An optimal subset of features is selected as input vector of SVM for training and recognition. Block recognized as light is eliminated. Experimental results demonstrate that SVM based method is more accurate and faster than other classifiers when used to eliminate the light interference.